A brief peek into the Framework for Adoption of AI in Banks from ECB

The publication ‘The rise of artificial intelligence: benefits and risks for financial stability by ECB’ gives a view with some deep fundamental insight into the hottest star around – AI. It touches on the conceptual aspects and more importantly acknowledges the evolving aspect of the scientific understanding of the technology itself & thus the ambiguity around risk & benefits that accrues in future. In a concise form- While AI efficacy into operations it also builds in operational risks too.

Overview

The publication does not want to be jump to conclusion as reflected in ‘any discussions of the benefits, risks and systemic consequences of AI are largely based on conjecture’.  It further states that the benefits and risks of AI depend on the individual use case. It believes that regulatory initiatives may be needed in case the current prudential framework current proves inadequate.

As it takes measured steps it brings to forefront the willingness of business to adopt AI for productivity gains ‘64% of businesses believe that AI will increase their productivity ‘, so also the concerns about technology dependence. It predicts that Banking is expected to be a large beneficiary as the overall add to the economy by generative AI  between USD 2.6 trillion and USD 4.4 trillion.

Conceptual Framework

Sets the record straight on AI as well which I quote verbatim ‘The term artificial intelligence may thus be a misnomer as it suggests “intelligence” yet in essence it is the outcome of a stochastic process that combines text based on probabilistic information the model does not fundamentally understand the underlying logic of the text’.

It deals with how ‘Foundation models form a knowledge base for generative AI’  describing that ‘These models are “trained” in a self-supervised manner on a vast amount of both structured (e.g. tables) and unstructured (images, sound, text) raw data with only minimal human intervention’. Whereby the model learns the “ground truth” in a generic way, as it recognises human language, recognition of objects and images & numerical input. Generative AI models can make use of the generic knowledge of foundation models also called ground truth. Generative AI models being on LLMs , thus eliminating the need for proficient coding skills to modify or use them. The performance of foundation models can be enhanced by providing additional training on task-related data.

It details a ‘Systematic overview of AI and sub-fields’ to facilitate understanding and distinguish different sub-fields of AI. There is no clear scientific taxonomy of AI and its sub-fields at this stage, but a possible classification, which is in line with scientific discussions and approaches – Point in time. Effectively it repeats the evolving nature of the emerging technology.

Efficacy to Business, the potential area of use & resultant Risks

AI brings both benefits and risks to the financial system bringing efficacy into operations while building the risk of operational risks. the overall impact will depend on how the challenges related to data, model development and deployment. This is further accentuated by very similar foundation models provided by a few suppliers.

The publications focus is a bit more on systemic risks- technological penetration & supplier concentration among others. Such may trigger when most financial institutions use the same or very similar foundation models provided by a few suppliers, this will have similar bias when it comes to decisions at micro level. Besides making the financial system more fragile. The playing field being uneven for smaller players as large players get the advantage. Like the progress of the tech, it is impossible to predict pros & cons of this supplier concentration.

To further top up the challenges broader macroeconomic and climate-related effects of AI as well as the moral and ethical aspects of the (mis-)use of AI need to be explored further. This is generic & a wait for specific  regulations around this is merited.

Conclusion

As practitioners build use case on potential areas the guidelines in the publication can be a framework. It must be kept in mind the meandering & dynamic nature of the technology & accordingly followed. Use cases which are substantial & potentially having operational efficacy issue can be taken up. Use cases should provision for the operational risk arising from the same. Operational efficacy is an issue around voluminous process like AML or Fraud to document intensive once.

References 

The rise of artificial intelligence: benefits and risks for financial stability